自适应预测反褶积

ADAPTIVE PREDICTIVE DECONVOLUTION

The deconvolution technique present in this paper is referred to a modification of the conventional least-square one, which normally begins with the determination of the locations where non-zero reflection coefficients lie. In practice, seismic data are categorized into two types according to which type of points, predictable or unpredictable, they are subordinate to. A predictive filter is then defined to make predicted errors minimun at the predictable points. In this way, the deconvolution process is completed. This paper pioneers the so-called " Location Deconvolution" by which reflection coefficients are correctly defined once the non-zero reflection coefficient locations are given. An adaptive iterative algorithm is also presented to succcsivcly estimate the predictable and unpredictable points for seismic data so as to determine the locations where non-zero reflection coefficients lie.The experiments on a thcoritical model show that the technique is suitable for non-white reflectivity sequences as well as the wavelets with various phases. The results from real data further confirm its effectiveness.